Period adaptation of real-time control tasks with fixed-priority scheduling in cyber-physical systems

Abstract

Abstract Long-lived, non-stop cyber-physical systems (CPS) are subject to evolutionary changes that can undermine the guarantees of schedulability that were verified at the time of deployment. At the same time, knowledge gleamed from extended periods of execution can be exploited to reduce the uncertainties that were inevitably presented in the system models that are used to define the temporal behaviours of the control tasks. In this paper we utilise this knowledge and present an adaptation method that actively extends the period of control tasks at run-time based on historical measurements. This can lead to lower power consumption or to the accommodation of increased computation resource demands from other components of the CPS. The method relies on online monitoring and model-based prediction to degrade control performance while having a minimal and acceptable impact on ongoing operations. Cloud-based computing is used to facilitate decision making and offload the local computation. We evaluate the effectiveness of the proposed method through control-scheduling co-simulation.